vm_machdep.c revision 879
1/*- 2 * Copyright (c) 1982, 1986 The Regents of the University of California. 3 * Copyright (c) 1989, 1990 William Jolitz 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * the Systems Programming Group of the University of Utah Computer 8 * Science Department, and William Jolitz. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * from: @(#)vm_machdep.c 7.3 (Berkeley) 5/13/91 39 * Utah $Hdr: vm_machdep.c 1.16.1.1 89/06/23$ 40 * $Id: vm_machdep.c,v 1.7 1993/11/25 01:31:02 wollman Exp $ 41 */ 42 43#include "npx.h" 44#include "param.h" 45#include "systm.h" 46#include "proc.h" 47#include "malloc.h" 48#include "buf.h" 49#include "user.h" 50 51#include "../include/cpu.h" 52 53#include "vm/vm.h" 54#include "vm/vm_kern.h" 55 56/* 57 * Finish a fork operation, with process p2 nearly set up. 58 * Copy and update the kernel stack and pcb, making the child 59 * ready to run, and marking it so that it can return differently 60 * than the parent. Returns 1 in the child process, 0 in the parent. 61 * We currently double-map the user area so that the stack is at the same 62 * address in each process; in the future we will probably relocate 63 * the frame pointers on the stack after copying. 64 */ 65int 66cpu_fork(p1, p2) 67 register struct proc *p1, *p2; 68{ 69 register struct user *up = p2->p_addr; 70 int foo, offset, addr, i; 71 extern char kstack[]; 72 extern int mvesp(); 73 74 /* 75 * Copy pcb and stack from proc p1 to p2. 76 * We do this as cheaply as possible, copying only the active 77 * part of the stack. The stack and pcb need to agree; 78 * this is tricky, as the final pcb is constructed by savectx, 79 * but its frame isn't yet on the stack when the stack is copied. 80 * swtch compensates for this when the child eventually runs. 81 * This should be done differently, with a single call 82 * that copies and updates the pcb+stack, 83 * replacing the bcopy and savectx. 84 */ 85 p2->p_addr->u_pcb = p1->p_addr->u_pcb; 86 offset = mvesp() - (int)kstack; 87 bcopy((caddr_t)kstack + offset, (caddr_t)p2->p_addr + offset, 88 (unsigned) ctob(UPAGES) - offset); 89 p2->p_regs = p1->p_regs; 90 91 /* 92 * Wire top of address space of child to it's kstack. 93 * First, fault in a page of pte's to map it. 94 */ 95 addr = trunc_page((u_int)vtopte(kstack)); 96 vm_map_pageable(&p2->p_vmspace->vm_map, addr, addr+NBPG, FALSE); 97 for (i=0; i < UPAGES; i++) 98 pmap_enter(&p2->p_vmspace->vm_pmap, (vm_offset_t)kstack+i*NBPG, 99 pmap_extract(kernel_pmap, ((int)p2->p_addr)+i*NBPG), 100 /* 101 * The user area has to be mapped writable because 102 * it contains the kernel stack (when CR0_WP is on 103 * on a 486 there is no user-read/kernel-write 104 * mode). It is protected from user mode access 105 * by the segment limits. 106 */ 107 VM_PROT_READ|VM_PROT_WRITE, TRUE); 108 pmap_activate(&p2->p_vmspace->vm_pmap, &up->u_pcb); 109 110 /* 111 * 112 * Arrange for a non-local goto when the new process 113 * is started, to resume here, returning nonzero from setjmp. 114 */ 115 if (savectx(up, 1)) { 116 /* 117 * Return 1 in child. 118 */ 119 return (1); 120 } 121 return (0); 122} 123 124#ifdef notyet 125/* 126 * cpu_exit is called as the last action during exit. 127 * 128 * We change to an inactive address space and a "safe" stack, 129 * passing thru an argument to the new stack. Now, safely isolated 130 * from the resources we're shedding, we release the address space 131 * and any remaining machine-dependent resources, including the 132 * memory for the user structure and kernel stack. 133 * 134 * Next, we assign a dummy context to be written over by swtch, 135 * calling it to send this process off to oblivion. 136 * [The nullpcb allows us to minimize cost in swtch() by not having 137 * a special case]. 138 */ 139struct proc *swtch_to_inactive(); 140volatile void 141cpu_exit(p) 142 register struct proc *p; 143{ 144 static struct pcb nullpcb; /* pcb to overwrite on last swtch */ 145 146#if NNPX > 0 147 npxexit(p); 148#endif /* NNPX */ 149 150 /* move to inactive space and stack, passing arg accross */ 151 p = swtch_to_inactive(p); 152 153 /* drop per-process resources */ 154 vmspace_free(p->p_vmspace); 155 kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 156 157 p->p_addr = (struct user *) &nullpcb; 158 splclock(); 159 swtch(); 160 /* NOTREACHED */ 161} 162#else 163void 164cpu_exit(p) 165 register struct proc *p; 166{ 167 168#if NNPX > 0 169 npxexit(p); 170#endif /* NNPX */ 171 splclock(); 172 swtch(); 173 /* 174 * This is to shutup the compiler, and if swtch() failed I suppose 175 * this would be a good thing. This keeps gcc happy because panic 176 * is a volatile void function as well. 177 */ 178 panic("cpu_exit"); 179} 180 181void 182cpu_wait(p) 183 struct proc *p; 184{ 185 186 /* drop per-process resources */ 187 vmspace_free(p->p_vmspace); 188 kmem_free(kernel_map, (vm_offset_t)p->p_addr, ctob(UPAGES)); 189} 190#endif 191 192/* 193 * Set a red zone in the kernel stack after the u. area. 194 */ 195void 196setredzone(pte, vaddr) 197 u_short *pte; 198 caddr_t vaddr; 199{ 200/* eventually do this by setting up an expand-down stack segment 201 for ss0: selector, allowing stack access down to top of u. 202 this means though that protection violations need to be handled 203 thru a double fault exception that must do an integral task 204 switch to a known good context, within which a dump can be 205 taken. a sensible scheme might be to save the initial context 206 used by sched (that has physical memory mapped 1:1 at bottom) 207 and take the dump while still in mapped mode */ 208} 209 210/* 211 * Move pages from one kernel virtual address to another. 212 * Both addresses are assumed to reside in the Sysmap, 213 * and size must be a multiple of CLSIZE. 214 */ 215void 216pagemove(from, to, size) 217 register caddr_t from, to; 218 int size; 219{ 220 register struct pte *fpte, *tpte; 221 222 if (size % CLBYTES) 223 panic("pagemove"); 224 fpte = kvtopte(from); 225 tpte = kvtopte(to); 226 while (size > 0) { 227 *tpte++ = *fpte; 228 *(int *)fpte++ = 0; 229 from += NBPG; 230 to += NBPG; 231 size -= NBPG; 232 } 233 tlbflush(); 234} 235 236/* 237 * Convert kernel VA to physical address 238 */ 239u_long 240kvtop(addr) 241 register void *addr; 242{ 243 vm_offset_t va; 244 245 va = pmap_extract(kernel_pmap, (vm_offset_t)addr); 246 if (va == 0) 247 panic("kvtop: zero page frame"); 248 return((u_long)va); 249} 250 251#ifdef notdef 252/* 253 * The probe[rw] routines should probably be redone in assembler 254 * for efficiency. 255 */ 256prober(addr) 257 register u_int addr; 258{ 259 register int page; 260 register struct proc *p; 261 262 if (addr >= USRSTACK) 263 return(0); 264 p = u.u_procp; 265 page = btop(addr); 266 if (page < dptov(p, p->p_dsize) || page > sptov(p, p->p_ssize)) 267 return(1); 268 return(0); 269} 270 271probew(addr) 272 register u_int addr; 273{ 274 register int page; 275 register struct proc *p; 276 277 if (addr >= USRSTACK) 278 return(0); 279 p = u.u_procp; 280 page = btop(addr); 281 if (page < dptov(p, p->p_dsize) || page > sptov(p, p->p_ssize)) 282 return((*(int *)vtopte(p, page) & PG_PROT) == PG_UW); 283 return(0); 284} 285 286/* 287 * NB: assumes a physically contiguous kernel page table 288 * (makes life a LOT simpler). 289 */ 290kernacc(addr, count, rw) 291 register u_int addr; 292 int count, rw; 293{ 294 register struct pde *pde; 295 register struct pte *pte; 296 register int ix, cnt; 297 extern long Syssize; 298 299 if (count <= 0) 300 return(0); 301 pde = (struct pde *)((u_int)u.u_procp->p_p0br + u.u_procp->p_szpt * NBPG); 302 ix = (addr & PD_MASK) >> PD_SHIFT; 303 cnt = ((addr + count + (1 << PD_SHIFT) - 1) & PD_MASK) >> PD_SHIFT; 304 cnt -= ix; 305 for (pde += ix; cnt; cnt--, pde++) 306 if (pde->pd_v == 0) 307 return(0); 308 ix = btop(addr-KERNBASE); 309 cnt = btop(addr-KERNBASE+count+NBPG-1); 310 if (cnt > (int)&Syssize) 311 return(0); 312 cnt -= ix; 313 for (pte = &Sysmap[ix]; cnt; cnt--, pte++) 314 if (pte->pg_v == 0 /*|| (rw == B_WRITE && pte->pg_prot == 1)*/) 315 return(0); 316 return(1); 317} 318 319useracc(addr, count, rw) 320 register u_int addr; 321 int count, rw; 322{ 323 register int (*func)(); 324 register u_int addr2; 325 extern int prober(), probew(); 326 327 if (count <= 0) 328 return(0); 329 addr2 = addr; 330 addr += count; 331 func = (rw == B_READ) ? prober : probew; 332 do { 333 if ((*func)(addr2) == 0) 334 return(0); 335 addr2 = (addr2 + NBPG) & ~PGOFSET; 336 } while (addr2 < addr); 337 return(1); 338} 339#endif 340 341extern vm_map_t phys_map; 342 343/* 344 * Map an IO request into kernel virtual address space. Requests fall into 345 * one of five catagories: 346 * 347 * B_PHYS|B_UAREA: User u-area swap. 348 * Address is relative to start of u-area (p_addr). 349 * B_PHYS|B_PAGET: User page table swap. 350 * Address is a kernel VA in usrpt (Usrptmap). 351 * B_PHYS|B_DIRTY: Dirty page push. 352 * Address is a VA in proc2's address space. 353 * B_PHYS|B_PGIN: Kernel pagein of user pages. 354 * Address is VA in user's address space. 355 * B_PHYS: User "raw" IO request. 356 * Address is VA in user's address space. 357 * 358 * All requests are (re)mapped into kernel VA space via the useriomap 359 * (a name with only slightly more meaning than "kernelmap") 360 */ 361void 362vmapbuf(bp) 363 register struct buf *bp; 364{ 365 register int npf; 366 register caddr_t addr; 367 register long flags = bp->b_flags; 368 struct proc *p; 369 int off; 370 vm_offset_t kva; 371 register vm_offset_t pa; 372 373 if ((flags & B_PHYS) == 0) 374 panic("vmapbuf"); 375 addr = bp->b_saveaddr = bp->b_un.b_addr; 376 off = (int)addr & PGOFSET; 377 p = bp->b_proc; 378 npf = btoc(round_page(bp->b_bcount + off)); 379 kva = kmem_alloc_wait(phys_map, ctob(npf)); 380 bp->b_un.b_addr = (caddr_t) (kva + off); 381 while (npf--) { 382 pa = pmap_extract(&p->p_vmspace->vm_pmap, (vm_offset_t)addr); 383 if (pa == 0) 384 panic("vmapbuf: null page frame"); 385 pmap_enter(vm_map_pmap(phys_map), kva, trunc_page(pa), 386 VM_PROT_READ|VM_PROT_WRITE, TRUE); 387 addr += PAGE_SIZE; 388 kva += PAGE_SIZE; 389 } 390} 391 392/* 393 * Free the io map PTEs associated with this IO operation. 394 * We also invalidate the TLB entries and restore the original b_addr. 395 */ 396void 397vunmapbuf(bp) 398 register struct buf *bp; 399{ 400 register int npf; 401 register caddr_t addr = bp->b_un.b_addr; 402 vm_offset_t kva; 403 404 if ((bp->b_flags & B_PHYS) == 0) 405 panic("vunmapbuf"); 406 npf = btoc(round_page(bp->b_bcount + ((int)addr & PGOFSET))); 407 kva = (vm_offset_t)((int)addr & ~PGOFSET); 408 kmem_free_wakeup(phys_map, kva, ctob(npf)); 409 bp->b_un.b_addr = bp->b_saveaddr; 410 bp->b_saveaddr = NULL; 411} 412 413/* 414 * Force reset the processor by invalidating the entire address space! 415 */ 416void /* XXX should be __dead too */ 417cpu_reset() { 418 419 /* force a shutdown by unmapping entire address space ! */ 420 bzero((caddr_t) PTD, NBPG); 421 422 /* "good night, sweet prince .... <THUNK!>" */ 423 tlbflush(); 424 /* NOTREACHED */ 425 while(1); /* to fool compiler... */ 426} 427